Why Fasting Insulin Is the Biomarker Almost Nobody Orders

Type 2 diabetes is almost always diagnosed on a glucose measurement: fasting plasma glucose, a two-hour oral glucose tolerance test, or an HbA1c. Diabetes Canada sets prediabetes at a fasting plasma glucose of 6.1–6.9 mmol/L, a two-hour value of 7.8–11.0 mmol/L, or an A1C of 6.0–6.4%, and diabetes at 7.0 mmol/L, 11.1 mmol/L and 6.5% respectively.^[1] Fasting insulin appears nowhere in those criteria.

That is a reasonable way to define a disease. It is a poor way to catch one early, because glucose is the last thing to go. The clearest demonstration comes from the Whitehall II study, which followed 6,538 British civil servants without diabetes and worked backwards from the 505 who went on to be diagnosed, using repeated measurements taken up to thirteen years before diagnosis.^[2]

The sequence is striking. HOMA-derived insulin sensitivity fell steeply throughout the five years before diagnosis. Beta-cell function rose between four and three years before diagnosis — from 85.0% to 92.6% — the pancreas compensating, working harder — and then collapsed to 62.4% by the time the diagnosis was made. Fasting glucose drifted upwards linearly for years and then turned sharply quadratic only three years out, climbing from 5.79 to 7.40 mmol/L. Two-hour post-load glucose did the same thing over the same three years, from 7.60 to 11.90 mmol/L.^[2]

In other words: insulin sensitivity degrades, insulin output rises to cover it, and glucose stays normal — until the compensation fails. A fasting glucose test looks for the failure. A fasting insulin test looks for the compensation.

Insulin is secreted by pancreatic beta cells in response to glucose, and its main fasting job is to restrain the liver from pouring glucose into the bloodstream overnight. If tissues respond poorly to insulin, the pancreas has to secrete more of it to achieve the same restraint. A fasting insulin drawn after an overnight fast is therefore a rough readout of how much insulin it currently takes to hold the line.

The usual way of formalising that is HOMA-IR, published in 1985, which combines fasting glucose and fasting insulin into a single number: fasting insulin (µIU/mL) × fasting glucose (mmol/L) ÷ 22.5, or ÷ 405 if glucose is in mg/dL. In the original paper the model's estimate of insulin resistance correlated well with the euglycaemic clamp (Rs = 0.88) — but also, almost as well, with the fasting insulin concentration alone (Rs = 0.81). The authors were candid about precision: the coefficient of variation was 31% for the insulin-resistance estimate, which they said “limits its use.”^[3]

Two of those facts deserve to sit next to each other. HOMA-IR is a good surrogate for a research-grade clamp. And it is, to a first approximation, fasting insulin with a glucose correction applied.

Common enough that the answer is uncomfortable. In six consecutive NHANES cycles covering 6,247 American adults aged 18 to 44 without diabetes or existing cardiovascular disease, 44.8% (95% CI 42.0–47.6) met a HOMA-IR threshold of 2.5 or above in 2007–2010, and 40.3% (36.4–44.2) did in 2015–2018. Insulin resistance travelled with everything else: among young adults with it, 56.6% were obese, 31.3% had hypertension and 16.0% had hypercholesterolaemia, against 14.7%, 14.7% and 7.0% in those without. The authors' own summary is the line worth remembering: four in ten young American adults have insulin resistance, and nearly half of those who do are not obese.^[4]

That echoes something Gerald Reaven argued in his 1988 Banting Lecture — that resistance to insulin-stimulated glucose uptake is present in the majority of people with impaired glucose tolerance or type 2 diabetes, and in roughly 25% of non-obese individuals with an entirely normal oral glucose tolerance test.^[5]

And the broader metabolic picture is no better. Using contemporary cut points for waist circumference, glucose, blood pressure, triglycerides and HDL, only 12.2% (95% CI 10.9–13.6) of American adults met every criterion for optimal metabolic health. Fewer than a third of normal-weight adults qualified; among adults with obesity, 0.5% did.^[6]

This is where an honest article has to turn around on itself, because there are four good reasons, and the enthusiasm online rarely mentions any of them.

The assay is not standardised. A work group convened by the American Diabetes Association measured 39 single-donor sera using ten commercial insulin methods from nine manufacturers and compared them against an isotope-dilution mass spectrometry reference procedure. Only four of the ten methods got 95% or more of individual results within 32% of the reference value. For seven of the ten methods, the bias against the reference exceeded 15.5% in somewhere between 36% and 100% of individual samples.^[7] A standardisation programme was launched to fix this,^[8] but laboratory guidance still cautions that not all manufacturers have recalibrated to the reference method, so results must be read against the specific assay that produced them.^[9] A fasting insulin of 9 at one lab is not reliably a fasting insulin of 9 at another, and it is certainly not comparable to a value from five years ago on a different platform.

There is no agreed cut-point. In a random Spanish population sample of 2,459 adults, the HOMA-IR threshold ranged from 3.46 if you simply take the 90th percentile of the population, down to 2.05 if you instead pick the value that best identifies people with clustered metabolic risk — 1.85 in non-diabetic men. All of those values sit between the 70th and 75th percentiles of the population.^[10] The number you are told is “high” depends entirely on which of those logics your lab or your clinician happens to use.

Laboratory medicine does not endorse it. The Association for Diagnostics & Laboratory Medicine's test-utilisation guidance is blunt: do not order insulin to diagnose diabetes, to monitor it, to distinguish type 1 from type 2, or to guide therapy — and specifically, that clinical assessment of insulin resistance using a fasting-insulin-based calculation such as HOMA-IR “is not recommended.”^[9] You can disagree with that position. You should not pretend it does not exist.

And the hard-outcome evidence is weaker than the physiology. A meta-analysis of seven prospective studies in 26,976 non-diabetic adults compared the highest with the lowest category of each measure. HOMA-IR was associated with all-cause mortality (RR 1.34, 95% CI 1.11–1.62) and, in the two studies reporting it, with cardiovascular mortality (RR 2.11, 95% CI 1.01–4.41). Fasting insulin on its own was not: RR 1.13 (95% CI 1.00–1.27, p = 0.058) for all-cause mortality and 1.40 (0.49–3.96) for cardiovascular mortality in the single study reporting it.^[11] A separate meta-analysis of seven studies in 36,343 participants found no significant association between HOMA-IR and incident stroke at all (RR 1.29, 95% CI 0.89–1.87), including in the non-diabetic subgroup.^[12] Both reviews flag small numbers of studies and substantial heterogeneity.

That is the honest shape of it. The physiology is solid and the epidemiology of prevalence is solid. The claim that the number itself predicts whether you will die is not.

Because insulin is awkward to measure, surrogate indices built only from routine chemistry have been proposed — most commonly the TyG index, derived from fasting triglycerides and fasting glucose. It correlates with adiposity and metabolic markers and shows moderate agreement with a clamp.^[13] But when 163 non-diabetic postmenopausal women underwent hyperinsulinaemic-euglycaemic clamps and oral glucose tolerance tests, TyG correlated with the clamp only modestly (r = −0.370) and performed worse on ROC analysis (AUROC 0.706) than an OGTT-derived index that includes insulin values (0.791). The authors concluded that indices containing insulin appeared more accurate, and that TyG needs wider validation before large-scale clinical use.^[14]

So the insulin-free shortcut does not currently outperform the insulin-containing measure. Both are approximations of a test almost nobody gets.

Here is the part that makes the test worth considering at all: what fasting insulin reflects responds quickly, and to unglamorous things.

In a randomised controlled trial, 5.1% ± 0.9% weight loss improved insulin sensitivity in adipose tissue, liver and muscle simultaneously, and improved beta-cell function — without any change in markers of inflammation. Larger losses of about 11% and 16% improved muscle insulin sensitivity and beta-cell function further, in a stepwise fashion.^[15] Five per cent is not an aspirational number; for an 85 kg adult it is a little over 4 kg.

The speed can be startling. In eleven people with type 2 diabetes placed on a 600 kcal/day diet, insulin suppression of hepatic glucose output improved from 43 ± 4% to 74 ± 5% — into the range of matched non-diabetic controls — and fasting plasma glucose normalised from 9.2 ± 0.4 to 5.9 ± 0.4 mmol/L, after one week. Liver triglyceride fell from 12.8% to 2.9% by week eight.^[16] Hepatic insulin sensitivity, which is largely what a fasting insulin is reading, is among the first things to recover.

Movement works too, independent of a scale. A meta-analysis of 50 studies found high-intensity interval training reduced insulin resistance compared with both non-exercising controls (SMD −0.49, 95% CI −0.87 to −0.12) and continuous training (SMD −0.35, −0.68 to −0.02), with a 0.19% fall in HbA1c and 1.3 kg of weight loss against controls.^[17]

And the most consequential trial of all never measured anyone's insulin to decide what to do. The Diabetes Prevention Program randomised 3,234 adults with elevated fasting and post-load glucose to placebo, metformin, or a lifestyle programme targeting 7% weight loss and 150 minutes of activity a week. Over an average 2.8 years, diabetes incidence was 11.0, 7.8 and 4.8 cases per 100 person-years — a 58% reduction with lifestyle (95% CI 48–66) and 31% with metformin (17–43). Only 6.9 people needed to take part in the lifestyle programme for three years to prevent one case of diabetes.^[18]

Which is the quiet point underneath this whole article. The actions a high fasting insulin should prompt — losing a modest amount of weight, moving more, sleeping properly, eating in a way you can sustain, and tracking what you actually eat rather than what you think you eat — are the actions that help whatever the number says. The test's value is motivational and chronological: it can tell you that something has started years before a glucose test will, which is exactly when the intervention works best. Peak Human's metabolic biomarker guide covers fasting insulin and HOMA-IR alongside glucose, HbA1c and triglycerides; our weight loss and nutrition pages cover the levers above.